00191369-01 - 第190页

6 Vision functions User Manual HS-50 6.2 PCB vision system Software-Version 5.01 Edition 01/99 188  .H\WR)LJ (1) Cam era fiel d of view (2) Se arch area ≤ camera field of v iew (the fiducia l is sea rched fo …

100%1 / 442
User Manual HS-50 6 Vision functions
Software-Version 5.01Edition 01/99 6.2 PCB vision system
187
 'HVFULSWLRQ R I) XQFWLRQV 
Before placement the location, skew and shear of the board is determined by the PCB vision sys-
tem using the position of the fiducials. Deviations from the setpoint values are then included in the
calculation of the placement positions of the components as corrections.
A board must have at least 2 fiducials if the system is to be able to detect deviations in the board
position and the skew of the board. The presence of 3 fiducials will furnish additional information
concerning compression or stretching of the board or of the board layout.
 6HTXHQFHRI)XQFWLRQV
Before a fiducial can be used for board recognition it must first be ’taught’ to the machine. In other
words, the fiducial structure parameters must have been saved in the PCB vision system for that
pattern.
The fiducial structure can be taught using the PCB vision camera mounted on the gantry and the
vision program. The vision analysis unit determines the significant fiducial structure parameters
using digital image processing methods.
Measurement takes place in two stages:
2-D pattern search (2-dimensional process) in the coarse grid and provisional determina-
tion of the fiducial coordinates
1-D pattern search (1-dimensional process) for a precise determination of the position of
the fiducials.
With the 2-D pattern search process the template window is divided into moxel areas. Moxels
(PRsaic pi[HOs) are pixel fields containing for example 16 x 16, 8 x 8 pixels and so on. The lower
the pixel count the higher the resolution and the lower the search speed.
6 Vision functions User Manual HS-50
6.2 PCB vision system Software-Version 5.01 Edition 01/99
188
.H\WR)LJ
(1) Camera field of view
(2) Search area
camera field of view (the fiducial is searched for in this area)
(3) Reference fiducial
(4) Moxel = pixel field, eg 16 x 16 pixels
(5) Template window (it contains the reference fiducial)
(6) Fiducial which is to be searched for
The template window is moved over the search area in moxel steps. The gray scale values of each
moxel of the reference fiducial are calculated at this time. This reduced data structure will contain
enough information on the coarse structure and position of the reference fiducial.
NOTE
The search window should be as small as possible in order to keep the speed of searching high.
On the other hand the window should be large enough to allow the fiducial to be identified without
ambiguity.
4
2
5
6
3
1
User Manual HS-50 6 Vision functions
Software-Version 5.01Edition 01/99 6.2 PCB vision system
189
The 1-D pattern search procedure is used for precisely determining the pattern and position of the
fiducial. The fiducial image is broken up into rows and columns and the gray scale values in each
row and column added up. The next diagram illustrates this process using a double cross.
.H\WR)LJ
(1) Fiducial
(2) Sum of the grey scale values in one column: column profile
(3) Sum of the grey scale values in one row: row profile
The position of the fiducial is precisely determined from the horizontal and vertical profiles.
After teaching, the fiducial structure parameters obtained are saved to the line computer.
The saved pattern is now tested. The gantry moves the PCB camera over the board to all 4
corners of the search area (worst case). During this test the vision system must re-identify the
fiducial four times.
Finally, the coordinates of each individual fiducial (at least two) are manually added to the NU
file or transferred from the CAD file to the NU file. In this way, the coordinates and fiducial struc-
ture parameters for the board are defined as a pattern in the system.
During the placement process, the fiducial parameters will be determined once again using the
1-D and 2-D image processing methods described above. The template window is moved
moxel by moxel over the search area searching for the best possible agreement of the gray
scale values of the reference and board search fiducials (correlation procedure). Maximum
correlation is attained when the reference and search fiducials agree.
Once the fiducial has been found, the 1-D pattern search process begins. The precise shape
1
2
3